Search Results (531)

Aptamer-based multiplexed proteomic platforms, especially the SOMAmer-based SomaScan assay, are widely used for large-scale discovery of circulating biomarkers relevant to human aging. This review summarizes 42 original research articles published from 2020 through 2026 in which aptamers or aptamer-derived biosensors were used to characterize aging-related biomarkers in human samples or clinically relevant human-disease contexts. The eligible literature falls into several thematic areas: whole-plasma and organ-specific proteomic aging clocks; inflammaging and senescence-associated secretory phenotype (SASP) markers; cardiovascular, metabolic, renal, hepatic, musculoskeletal and neurodegenerative biomarker panels; and aptasensor platforms for detection of individual analytes. Only a small number of studies have compared aptamer- and antibody-based platforms in the same specimens; we tabulate these and show that median between-platform agreement is low to moderate, which constrains the pooling of findings across technologies. We also make explicit an interpretive point that is usually left implicit: because proteomic clocks are trained against chronological age, their correlation with chronological age measures fit to the training target rather than biological validity, and the informative quantity is the residual age gap. In the reviewed literature, SomaScan-based studies are concentrated in cardiovascular, neurodegenerative, frailty, and proteomic aging-clock research, whereas de novo SELEX campaigns targeting aging-specific epitopes and longitudinal human validation of wearable aptasensors were not identified. The main barriers to translation are cross-platform discordance, limited replication across ancestries, under-reported pre-analytical variability, cost, and the research-use-only status of most assays. Full article

Power boiler tubes are exposed to severe operating conditions that lead to wall thinning and material degradation. Reliable assessment of tube wall thickness is therefore essential for ensuring safe and efficient boiler operation. This exploratory laboratory study investigates the applicability of active thermography combined with analytical and machine learning (ML) approaches for non-contact wall thickness estimation in power boiler tubes. Experimental investigations were performed on a single boiler tube specimen with artificially introduced wall-thickness reductions. Thermal responses were recorded using an infrared camera under both heating and cooling excitation conditions. Based on the acquired thermographic data, analytical models and machine learning algorithms were developed to estimate tube wall thickness. The machine learning approach was implemented using Random Forest and Support Vector Regression models and compared with conventional analytical modeling techniques. For separately analyzed and relatively homogeneous measurement series, the machine learning models produced lower descriptive errors than the analytical models, with the estimated three-RMSE error envelope decreasing from 0.51 mm to 0.17 mm. However, when heating and cooling datasets were aggregated, the analytical models achieved lower root mean square error values and demonstrated greater stability than the machine learning methods. These findings indicate that model performance strongly depends on the size, characteristics, and homogeneity of the available training data. Owing to the limited number of independent measurement series, the reported results should be interpreted as a small-sample feasibility assessment rather than as evidence of the general superiority of machine learning modeling. The results support the potential of active thermography for non-contact assessment of boiler tube wall thickness under controlled laboratory conditions. Further validation using additional specimens, grouped cross-validation, and physics-based synthetic data is required before the methodology can be considered for industrial implementation. Full article

As the scaling of silicon-based devices approaches physical limits, two-dimensional transition-metal dichalcogenide field-effect transistors (2D TMD FETs) have emerged as promising candidates for logic devices in the post-Moore era. However, their design optimization relies heavily on computationally intensive TCAD simulations, thereby limiting efficient exploration of multidimensional parameter spaces. This paper proposes an efficient automated design framework for 2D TMD FETs under small-sample conditions and validates it using a monolayer MoS₂ FET as a case study. The framework integrates device design, physics-based simulation, performance prediction, and inverse design, establishing a bidirectional mapping between device parameters and electrical performance. Target-driven closed-loop optimization is achieved through TCAD-based feedback validation. Results demonstrate that, using a dataset comprising 300 TCAD samples, the forward model achieves an average coefficient of determination (${\mathrm{R}}^2$) of 0.9503. TCAD revalidation of the inverse-designed devices yields an average mean absolute error (MAE) of 0.0464 and an average mean absolute percentage error (MAPE) of 5.46% for performance metrics. Regarding computational efficiency, while a single TCAD simulation takes approximately 25 to 50 min, the trained model performs inference in under 50 ms, achieving a speedup of at least $3\times {10}^4$ during the inference phase. Accounting for the generation of the 300 TCAD samples and the training of both forward and inverse models, the framework’s one-time computational cost ranges from 160.27 to 285.27 h. Once the cumulative number of design tasks exceeds approximately 342 to 385, the total computational cost falls below that of direct TCAD simulation, with the computational advantage becoming increasingly significant as the number of tasks grows. Consequently, this method is highly suitable for large-scale parameter sweeps, device screening, and multi-objective, high-frequency design iterations. It drastically reduces repetitive TCAD calls, offering a scalable solution for the efficient, automated design of 2D TMD FETs. Full article

A post-flight analysis of unmanned aerial vehicle (UAV) imagery has the potential to result in a delay in the inspection of source water. This delay can occur when visible debris or changes in the water surface necessitate a prompt response. The present study does not evaluate in-flight operation; rather, it presents a bench-level feasibility assessment of two deployment tasks—broad two-class screening and close-range debris classification—using lightweight YOLO detectors on an edge-AI camera in a host-fed configuration that approximates the timing constraints of a future UAV workflow. The YOLOv8, YOLO11, and YOLO26 models were lightweighted through structural pruning (YOLOv8) or architecture scaling (YOLO11 and YOLO26). These models were then refined through a process of fine-tuning, exported to the camera, and evaluated in terms of several metrics. The metrics encompassed training-environment accuracy, the accuracy of device-returned outputs, round-trip latency, and snapshot-based operating-load estimates. The dataset under consideration is extensive, comprising 4813 training images and 575 validation images, accompanied by 13,051 and 1615 annotations, respectively. The depth-pruned YOLOv8s variant demonstrated a significant reduction in mean round-trip latency, from 426.87 milliseconds to 231.58 milliseconds (45.75%), while the mAP@0.5 metric exhibited a decrease from 0.7018 to 0.6650, and the mAP@0.5:0.95 metric demonstrated a decline from 0.5433 to 0.5290. A class-level analysis reveals that aggregate accuracy is primarily influenced by the weaker floating-debris class, whose AP@0.5 ranges from 0.29 to 0.46, in contrast to the 0.82 to 0.94 range observed for pond/reservoir. In comparison to a matched baseline that was trained for an equivalent number of epochs with the sampler disabled, debris-biased sampling contributes 1.5 ± 0.6 mAP@0.5 points for YOLO11 and 3.6 ± 0.2 points for YOLO26 across three seed-matched pairs. The primary effect of this method is to increase floating-debris recall by 4.7–5.9 percentage points, with a concomitant small reduction in precision. The latency reduction increased the broad-inspection rate by 1.85×, provided approximately 195 milliseconds of idle margin within a 1-hertz cycle, and increased the paired far/near rate by 1.59× with two models resident on the camera. Three-seed repetitions of compact-model fine-tuning yielded 0.6717 ± 0.0033 and 0.6290 ± 0.0028 mAP@0.5. These results express detector compression in terms of operational monitoring capacity rather than model-size reduction alone, while also showing that compression by itself does not resolve the weak-class limitation that governs source-water inspection accuracy. Full article

Banking crises are a persistent threat to macroeconomic stability in emerging markets, where conventional econometric monitoring frameworks often fail to capture non-linear macro-financial relationships. This paper examines whether machine learning algorithms can improve the detection of banking crisis risk in Nigeria compared to standard logistic regression. We compare the performance of Random Forest, Support Vector Machine (SVM), and Extreme Gradient Boosting (XGBoost) against logistic regression using annual data from the African Financial Crises dataset (1954–2014). Resampling is only implemented on the training set to overcome the infrequency of crisis events. Performance on models is assessed based on accuracy, precision, recall, F1-score, and the area under the receiver operating characteristic curve (AUC) in a rigorous out-of-time validation setting. Our findings indicate that tree-based ensemble models outperform logistic regression on the test set: XGBoost achieves the best generalization performance (AUC = 1.0; F1 = 0.95 in non-crisis, 0.80 in crisis), whereas Random Forest has the highest cross-validated F1-score on the training set. The most important variables are exchange rate volatility, inflation, and indicators of systemic crisis. The most significant crisis indicators are, however, seen in crisis years, which means that the annual data do not provide much lead-time to detect the crisis. These results should be taken with caution because of the small sample size and the limited number of crisis observations during the test period. Altogether, machine learning models have potential as additional tools to monitor banking crises in Nigeria, though at the moment they are not fully operational as policy instruments. Full article

Radar-based Quantitative Precipitation Estimation (QPE) in semi-arid Mediterranean climates is critically challenged by systematic underestimation of shallow precipitation, yet gauge–radar merging frameworks tailored to such environments remain poorly evaluated. This study develops and assesses a merging pipeline for Cyprus, combining X-band polarimetric observations from the Paphos and Larnaca operational radar network with accumulations from a 50-station rain gauge network across 11 rainfall events spanning the 2024 wet season (January and November–December 2024). Four approaches were evaluated: raw radar mosaic, global mean field bias (MFB) correction, spatially varying local inverse distance weighting (IDW) bias correction assessed through leave-one-out cross-validation (LOOCV), and a Random Forest (RF) machine-learning retrieval trained on polarimetric, geometric, and orographic predictors and evaluated through leave-one-event-out cross-validation (LOEO-CV). Raw radar exhibited severe and highly variable underestimation, with station-level bias factors ranging from 1.4 to 200×. Global MFB correction removed systematic offset but, as a single spatially uniform scalar, could not improve spatial correspondence; it was beneficial only where the bias field was spatially coherent. Local IDW correction provided cross-validated reduction in RMSE for most events (commonly 40–53%), but this improvement reflected removal of mean bias rather than recovery of spatial pattern: only 17 January 2024 combined RMSE reduction (24.07 mm to 11.31 mm) with genuine spatial skill (leave-one-out r = 0.850, bias-field coherence r = 0.742), while several events improved in RMSE yet retained near-zero spatial correlation, and 30 and 31 January degraded outright. These results characterise the limits of distance-weighted (IDW) interpolation specifically; whether geostatistical estimators incorporating topographic external drift can restore spatial skill where the present gauge network constrains the bias field remains to be tested. When re-evaluated on the same rainy matched-pair set (N = 2378), the Random Forest reduced 10 min RMSE by only 2.6% relative to the best classical Z-R estimator (from 10.38 mm to 10.10 mm) and reduced the systematic bias from −5.06 mm to −4.25 mm, but did not improve point-to-point spatial correspondence (r ≈ 0), indicating that this mean-regression Random Forest provides effective bias-correction skill without spatial-correspondence skill, leaving the fundamental representativeness gap between CAPPI sampling and gauge point measurements unresolved. Three pre-conditions for local bias correction skill are identified as empirical diagnostics under the sample conditions of this study: a minimum of approximately 40 contributing gauges, a spatially coherent bias field, and a moderate bias range. A formal bootstrap or resampling-based uncertainty estimate for these indicators was not attempted, because eleven events constitute too small a sample for stable resampling statistics; the per-event relationships between the number of contributing gauges, the bias-factor range, the bias-field spatial autocorrelation, and the LOOCV error are therefore presented as the empirical basis for these diagnostic indicators, which should be refined and tested for statistical robustness as longer event records become available. These findings demonstrate that the suitability of spatial merging can be diagnosed from network and bias field properties prior to correction, and that machine-learning retrieval offers complementary value through systematic bias removal where spatial interpolation fails. Probabilistic merging frameworks, denser gauge networks, and ML approaches that explicitly target spatial correspondence are identified as priority developments for eastern Mediterranean QPE. Full article

To address the challenges of ambiguous regional feature identification, high digital modeling costs, and limited regional adaptability of general-purpose AIGC tools in the micro-renewal of Western Sichuan Linpan, this study proposes a typology-constrained small-sample learning framework. A structured dataset of 80 high-quality, semantically annotated images was constructed to extract typological features, including spatial layouts, architectural forms, and environmental relationships. Based on Stable Diffusion, the framework integrates LoRA fine-tuning for style adaptation and ControlNet for structural control. To constrain the number of trainable parameters under the limited-data setting, LoRA rank control, dropout regularization, and early stopping were incorporated into the training procedure. Under the fixed 80-image dataset and experimental configuration, the complete framework achieved lower FID values and higher CLIP and expert-evaluation scores than the prompt-only baseline. Compared with Scheme A, the FID decreased by 22.7% (from 284.52 to 220.00), while the CLIP Score increased by 18.3% (from 0.224 to 0.265). Under the evaluated configuration, the complete framework showed closer correspondence with the specified Linpan architectural characteristics and spatial conditions than the prompt-only baseline. The results support the feasibility of translating architectural typological knowledge into semantic and geometric conditioning signals under the evaluated limited-data setting. However, the influence of training-set size on model stability and generation performance was not examined and requires further investigation. Full article

Background: Retinoblastoma (RB) is a rare pediatric malignancy characterized by small sample sizes and low event rates, where conventional association-driven feature selection may lead to unstable models, overadjustment, and limited generalizability. However, existing survival prediction studies lack a careful treatment of feature selection that accounts for underlying causal structure. Objectives: To develop and validate a causality-guided machine learning model for RB survival prediction by jointly incorporating survival time and survival status as outcome variables. Methods: We analyzed 1015 RB patients from the SEER database (1975–2020). A causality-informed feature selection framework was developed to address the challenges of rare-disease data. Specifically, candidate variables were evaluated through a three-step evidence-integration process: (1) univariate Cox proportional hazards (CPH) analysis for initial statistical screening; (2) causal structure learning using the PC algorithm on the variables retained from Step 1 to construct a directed acyclic graph (DAG) and exclude structurally inappropriate variables (colliders or descendants of the outcome); and (3) LASSO-based feature screening performed independently on the full set of candidate variables. The final features were obtained by taking the intersection of the variables retained from Step 2 and Step 3. Survival models were then trained using the selected features, with model comparison performed as a secondary step. Results: The proposed framework consistently identified four structurally and prognostically robust predictors—laterality, “SEER historic stage A”, “RX Summ”, and sequence number—through this evidence-integration process. Compared with conventional approaches, the causality-informed framework reduced the feature set while improving model stability and interpretability. Notably, compared with LASSO-only selection, which retained a larger set of variables, the causality-informed approach yielded a more parsimonious feature set with improved predictive performance, suggesting reduced overfitting in a low-event setting. Survival models trained on this refined feature set demonstrated reliable predictive performance, with the random survival forest achieving the highest discrimination (C-index = 0.739). Importantly, the selected predictors aligned with clinically plausible pathways in the learned DAG, supporting their causal relevance. Conclusions: This study demonstrates that incorporating causal structure into feature selection provides a more reliable and interpretable foundation for survival modeling in retinoblastoma. Rather than focusing on algorithmic comparison alone, our findings highlight that careful, causality-informed feature selection is critical for improving robustness in rare-disease prediction tasks. This framework may serve as a generalizable methodological template for other rare clinical settings prone to spurious associations. Full article

Background/Objectives: Artificial intelligence (AI) has emerged as a transformative tool in oral medicine, where it holds significant promise for enhancing the diagnosis, treatment, and management of oral cancer. Our team aimed to develop a new tool, capable of diagnosing oral cancer utilizing the capabilities of AI. Methods: The task we aim to solve from computer vision’s perspective is an object detection task. Under this context, a detection is essentially a bounding box drawn around an oral lesion accompanied by the disease’s description. To solve the task at hand, we collected and annotated a wide set of images which were used for training our model. Specifically, we used 205 images of Oral Squamous Cell Carcinoma (OSCC). Following common practice, 80% of the total images were allocated for training, 10% for validation, and 10% for testing. The training set was used to optimize the model’s parameters across multiple iterations. The validation set served to prevent overfitting during training and to guide hyperparameter tuning. Lastly, the test set was used for evaluation of data that had not been previously seen by the model and had not influenced any decisions regarding its architecture or hyperparameters. Moreover, during evaluation, we supplemented the test set by adding 100 images of healthy mucosa to examine whether the model generated false positives on healthy tissue. To broaden the dataset’s coverage, we generated synthetic images by applying data augmentation techniques such as random rotation, scale and noise injection. The model’s architecture was based on YOLO11, which is a widely spread neural network architecture known for its balance between efficiency and performance, used in object detection tasks. Results: The model’s detections were accompanied by a confidence measure, which was used to filter out those with low confidence, and one could choose a lower threshold for maximizing precision or a higher threshold for maximizing recall. Among images that correspond to oral cancer (oral squamous cell carcinoma), the model achieved 59% precision and 41% recall on the validation set and 56% precision and 42% recall on the test set. The limitations of this study include the single institutional design and the relatively small sample size. The number of images used for model training was relatively small (205 images of oral squamous cell carcinoma), which may limit the generalizability of the findings. The main limitation is that the model distinguishes oral squamous cell carcinoma from healthy mucosa. In routine clinical practice, however, the diagnostic challenge is to differentiate oral cancer from a variety of benign and potentially malignant disorders that may present with similar clinical features. The inclusion of other oral lesions is planned in future studies. Conclusions: The efficiency of our AI application may be considered as encouraging, taking the pilot nature of the study into consideration. More clinical photos and better training of the model may lead to better precision and recall, enabling its inclusion in standard clinical practice. Larger multicenter datasets will be required for clinical implementation. AI-driven tools can assist in risk stratification, helping clinicians determine the best treatment plans by analyzing patient data and predicting the likelihood of recurrence or metastasis. AI’s potential extends beyond diagnosis and treatment; it also contributes to monitoring patient outcomes. As research and technology evolve, AI’s role in oral cancer is likely to become increasingly indispensable. Full article

Vehicle-ejected debris detection is a practical but insufficiently studied problem in intelligent traffic enforcement. Unlike static road litter, objects thrown from moving vehicles are usually small, irregular, transient, and easily confused with road textures, shadows, lane markings, and light reflections. In current traffic management, such violations still rely heavily on manual video review or offline inspection, while task-specific datasets and edge-deployable detection solutions remain limited. To address this gap, this study constructs a vehicle-ejected debris dataset containing 4328 annotated image samples collected from real road scenarios. The dataset covers urban and suburban roads, daytime and nighttime illumination, near-range and distant small-object cases, and hard negative samples. To meet the coupled requirements of vehicle-mounted small-object detection and edge-side INT8 deployment, this study develops a hardware-aware lightweight detection framework based on YOLOv8m. The original CSPDarknet backbone is replaced with the convolutional variant of MobileNetV4 to reduce feature-extraction cost, while a scale-specific Channel Alignment Module is inserted between the heterogeneous MobileNetV4 backbone and the YOLOv8m PANet neck to preserve multi-scale feature compatibility. The alignment module uses only BPU-friendly convolution, batch normalization, and activation operations, thereby avoiding deployment-unfriendly operators while maintaining compatibility with INT8 quantization and edge acceleration. The trained FP32 model is quantized to INT8 and deployed on the RDK X5 BPU using the Horizon OpenExplorer toolkit. Experimental results and repeated-seed validation show that the proposed model achieves a consistent accuracy–efficiency advantage on the constructed dataset. In a representative run, the proposed model obtains 93.1% mAP₅₀, while reducing the number of parameters from 25.9 M to 13.1 M and GFLOPs from 78.9 to 39.6 compared with the YOLOv8m baseline. After INT8 deployment, the model reaches 112.6 FPS on the RDK X5 platform with only a minor accuracy decrease. These results indicate that the proposed framework can serve as a practical edge-deployable perception module for real-time vehicle-ejected debris monitoring under vehicle-mounted traffic-enforcement scenarios. It should be noted that this work focuses on single-frame debris detection, while event-level ejection verification, temporal consistency analysis, offending-vehicle attribution, and enforcement decision-making remain beyond the scope of this study. Full article

The resilience pads of sleeper floating tracks (STEDEF) are key components that absorb shock loads and vibrations induced by train traffic. Currently, under the Korean domestic guidelines for track facility performance evaluation, one sample per 500 m is collected from the field, and the static spring stiffness is assessed through laboratory testing. However, this approach requires nighttime track possession, incurs significant manpower and cost, and provides limited reliability because the condition of an entire section is inferred from a small number of samples. Therefore, this study proposes an impact-hammer–FRF-based method for evaluating the spring stiffness of resilience pads without pad extraction. Field impact-hammer tests were conducted to identify the dominant first-mode natural frequency of the track-support system. Configuration-specific finite element models were then used to derive frequency–stiffness relationships for the investigated STEDEF configurations. The novelty of the proposed method lies in converting the local first-mode frequency measured in situ into a static-equivalent stiffness index that can be directly compared with the maintenance reference value used in the Korean inspection framework. The finite element model reproduced the mean natural frequencies of the reference configurations with differences of 0.08–3.61%. Based on the configuration-specific relationships, estimation equations were developed to support in situ screening of resilience-pad stiffness at the measured locations. Full article

Motor impairments and limitations in functional performance are common in children with autism spectrum disorder, restricting participation in daily activities. This study aimed to compare the effectiveness of the MIG Program with conventional physical therapy in the development of socio-communicative motor skills and the achievement of functional goals. A mixed-methods randomized clinical trial was conducted with children with autism spectrum disorder aged 6 to 12 years (mean 8.73 ± 1.95; support levels 1 and 2), recruited from rehabilitation clinics in southeastern Brazil. Participants were randomly assigned to the MIG Program, which integrates contextualized functional motor training with narrative grammar strategies and the use of a therapeutic vest, or to conventional physical therapy based on traditional motor approaches. Primary outcomes included fundamental motor skills and functional goal attainment, while secondary outcomes were balance, gross and fine motor skills, and socio-communicative abilities. The RCT protocol was registered in the Brazilian Clinical Trials Registry (RBR-76pk39r), in 21 October 2025. The MIG Program was associated with greater improvements in fundamental motor skills and functional goal attainment compared to conventional physical therapy, with effects maintained at follow-up, as well as with more favorable trends in balance and communication outcomes; however, no clear differences were observed in gross and fine motor skills. Qualitative findings suggested increased engagement, autonomy, and participation in the MIG group. Overall, these preliminary findings indicate that the MIG Program may be a promising approach for supporting functional outcomes in children with autism spectrum disorder, although the results should be interpreted with caution given the small sample size and the number of outcomes assessed. Full article

Few-shot node classification aims to effectively predict new class nodes using only a small number of labeled samples, which is an important research direction in graph data mining. Existing self-training-based few-shot node classification methods are constrained by the bias and local optima of a single-teacher model. Meanwhile, multiple-teacher models often suffer from knowledge conflicts and redundant information, which degrade distillation efficiency and model generalization performance. To address these issues, we propose a multi-teacher knowledge distillation method with reinforcement learning for few-shot node classification (MTKD-RL). This framework is composed of a multi-teacher distillation network and a multi-teacher weight optimization module to deliver complementary supervision information from multiple perspectives. The reinforcement learning agent dynamically assigns adaptive weights to different teachers based on their prediction performance and the discrepancy between teacher and student models, which greatly enhances pseudo-label quality and distillation performance. Experiments on nine graph network datasets demonstrate that our method achieves consistent accuracy improvements ranging from 1.4% to 6.2% in few-shot node classification. Full article

Different mixtures of multimodal training data significantly impact the performance of multimodal large language models, and manually tuning data mixtures is inefficient, computationally expensive, and frequently suboptimal because of complex, nonlinear inter-modal interactions. How to determine data-mixture hyperparameters in an efficient and principled manner becomes the bottleneck for progress in the field. This study establishes a scalable, learnable framework, DMPredictor, that treats multimodal data-mixture design as a regression-based hyperparameter-optimization problem and automates the selection of effective training data mixtures. DMPredictor is trained on data mixture samples derived from hundreds of small proxy models (2M parameters), each of which is trained on 1B tokens sampled using different data mixtures. The framework incorporates alignment-aware smoothing and quality-reweighting, enabling diverse exploration of the multimodal data mixture space while avoiding distribution collapse. DMPredictor produces accurate performance forecasts and identifies nearly optimal data mixtures. The predicted optimal mixture surpasses human-designed baselines on diverse benchmarks, achieving +2.7% on MMMU, +6.4% on TextVQA, and +195.2 on MME. Moreover, the mixture optimization complexity is largely reduced by small proxies and a small number of tokens. The proposed approach offers a robust, computationally efficient pathway for optimizing mixtures of multimodal training data, addressing the critical challenge of training data heterogeneity. Full article

People living with neurological disorders frequently experience gait impairments that substantially reduce mobility, independence, and quality of life. This pilot study aimed to evaluate the feasibility, safety, and preliminary functional outcomes of integrating the EksoNR robotic exoskeleton (Ekso Bionics, San Rafael, CA, USA) into outpatient neurorehabilitation practice in individuals with chronic neurological impairments. Over an eight-month period, five participants with heterogeneous neurological conditions (two spinal cord injuries, one cerebellar ataxia, one ischemic stroke, and one spastic paraparesis) completed a four-week robotic gait training program consisting of 15 intervention sessions. Functional outcomes were assessed before and after the intervention using standardized clinical tests. Cardiovascular endurance was evaluated using the 6-Minute Walk Test (6MWT), while physical and psychological well-being were assessed with the Functional Independence Measure (FIM) and the Barthel Index, in addition to the WHO Quality of Life (WHOQOL) and EQ-5D-5L questionnaires. Mobility and balance were evaluated using the Timed Up and Go (TUG), Berg Balance Scale (BBS), Tinetti Performance-Oriented Mobility Assessment (POMA), and Walking Index for Spinal Cord Injury II (WISCI II), where applicable. In addition, device-recorded gait parameters, including step count, step length, walking distance, and walking duration, were analyzed. Significant improvements were observed in several device-derived gait parameters, including the number of steps performed with the exoskeleton (p < 0.001), step length (p = 0.003), walking distance (p = 0.002), and walking duration (p < 0.05). Significant improvements were also identified in balance performance (BBS: p = 0.006; Tinetti POMA: p = 0.001), cardiovascular endurance (6MWT: p = 0.017), and EQ-5D-5L scores (p = 0.038). Functional independence measures (FIM and BI), TUG performance, and WHOQOL domains did not demonstrate statistically significant changes. No serious adverse events or device-related injuries occurred during the intervention period. Due to the small and clinically heterogeneous sample, these findings should be interpreted as preliminary exploratory results. Nevertheless, the study supports the feasibility and potential clinical utility of EksoNR-assisted gait rehabilitation and provides a basis for larger controlled investigations. Full article